
DiffCut achieves state-of-the-art unsupervised zero-shot semantic segmentation by leveraging the semantic richness of diffusion UNet encoder features within a flexible recursive graph partitioning framework.


coremsg

diffcut-unsupervised-zero-shot-semantic-segmentation-using-diffusion-features-and-recursive-normalized-cut


DiffCut: Unsupervised Zero-Shot Semantic Segmentation Using Diffusion Features and Recursive Normalized Cut


title_rewrite


To mitigate the objective misalignment issue in zero-shot semantic segmentation, AlignZeg employs a comprehensive approach including Mutually-Refined Proposal Extraction, Generalization-Enhanced Proposal Classification, and Predictive Bias Correction, leading to significant improvements in unseen class recognition.


alignzeg-mitigating-objective-misalignment-for-improved-zero-shot-semantic-segmentation


AlignZeg: Mitigating Objective Misalignment for Improved Zero-shot Semantic Segmentation



Enhancing semantic alignment and generalization in zero-shot semantic segmentation through a language-driven visual consensus approach.


language-driven-visual-consensus-for-zero-shot-semantic-segmentation-enhancing-alignment-and-generalization


Language-Driven Visual Consensus for Zero-Shot Semantic Segmentation: Enhancing Alignment and Generalization
